广义回归神经网络预测加筋土支挡结构高度

土工合成材料加筋支挡结构（Geosythetics-Reinforced Retaining Wall, 简称GRW）设计方法主要是建立在似粘聚力理论基础之上的半经验设计法。由于土性及加筋机理的复杂性，常常要对它们进行人为假定，导致计算结果差强人意。神经网络方法与传统方法的不同之处在于不需要主观假定，而是模拟人脑思维，通过数据样本的学习来获得预测结果。引入神经网络技术来预测加筋土支挡结构的设计高度是一种新尝试。由于本问题具有样本容量非常有限、影响因素复杂多样的特点。因此，采用适用于稀土样本数据的广义回归网络（General Regression Neural Network)来预测加筋土支挡结构设计高度。基于MATLAB神经网络工具箱及文献[1]的挡墙离心模型试验结果，建立了一个可用于加筋支挡结构设计高度预测的GRNN网络。通过对足尺试验，实际工程及模型试验结果的检验，表明网络的学习是成功的，具有一定指导意义。     

Estimating width of the stable channels using multivariable mathematical models

Predicting behavior and the geometry of the channels and alluvial rivers in which the erosion and sediment transport are in equilibrium is one of the most important topics in river engineering. Various researchers have proposed empirical equations to estimate stable river width (W). In this research, empirical equations were examined and tested with a comprehensive available data set consisting of 1644 points collected from 29 stable rivers in various parts of the world. The data set covers a wide range of flow conditions, river geometry, and bed sediments. This data set is classified in two groups (W < 600 m and W ≥ 600 m) for presenting the new models. The new linear and nonlinear multivariable equations were fitted to these two groups, and the best models were selected by preliminary tests and diagnostic determined for each group. The determination coefficient of these models ranged from 0.87 to 0.96. The results show that the models presented in this paper are more accurate with respect to the previously presented models. In the second part, “Artificial neural networks,” perceptron was used and a new methodology for estimating stable channel width was developed. Comparison of the statistical methods presented in this paper and the results of perceptron neural network revealed preferential recent method.     

A neural network approach for attenuation relationships: An application using strong ground motion data from Turkey

This paper presents an application of neural network approach for the prediction of peak ground acceleration (PGA) using the strong motion data from Turkey, as a soft computing technique to remove uncertainties in attenuation equations. A training algorithm based on the Fletcher–Reeves conjugate gradient back-propagation was developed and employed for three sample sets of strong ground motion. The input variables in the constructed artificial neural network (ANN) model were the magnitude, the source-to-site distance and the site conditions, and the output was the PGA. The generalization capability of ANN algorithms was tested with the same training data. To demonstrate the authenticity of this approach, the network predictions were compared with the ones from regressions for the corresponding attenuation equations. The results indicated that the fitting between the predicted PGA values by the networks and the observed ones yielded high correlation coefficients (R²). In addition, comparisons of the correlations by the ANN and the regression method showed that the ANN approach performed better than the regression. Even though the developed ANN models suffered from optimal configuration about the generalization capability, they can be conservatively used to well understand the influence of input parameters for the PGA predictions.     

Neural network based prediction of ground surface settlements due to tunnelling

Ground surface settlement due to tunnel excavation varies in magnitude and trend depending on several factors such as tunnel geometry, ground conditions, etc. Although there are several empirical and semi-empirical formulae available for predicting ground surface settlement, most of these do not simultaneously take into consideration all the relevant factors, resulting in inaccurate predictions. In this study, an artificial neural network (ANN) is incorporated with '113' of monitored field results to predict surface settlement for a tunnel site with prescribed conditions. To achieve this, a standard format (a protocol) for a database of monitored field data is first proposed and then used for sorting out a variety of monitored data sets available in KICT. Using the capabilities of pattern recognition and memorization of the ANN, an attempt is made to capture the rich physical characteristics smeared in the database and at the same time filter inherent noise in the monitored data. Here, an optimal neural network model is suggested through preliminary parametric studies. It is shown that preliminary studies for generating an optimal ANN under given training data sets are necessary because no analytical method for this purpose is available to date. In addition, this study introduces a concept of relative strength of effects (RSE) [Yang Y, Zhang Q. A heirarchical analysis for rock engineering using artificial neural networks. Rock Mechanics and Rock Engineering 1997; 30(4): 207–22] in sensitivity analysis for various major factors affecting the surface settlement in tunnelling. It is seen in some examples that the RSE rationally enables us to recognize the most significant factors of all the contributing factors. Two verification examples are undertaken with the trained ANN using the database created in this study. It is shown from the examples that the ANN has adequately recognized the characteristics of the monitored data sets retaining a generality for further prediction. It is believed that an ANN based hierarchical prediction procedure shown in this paper can be further employed in many kinds of geotechnical engineering problems with inherent uncertainties and imperfections.     

Spatial and temporal monthly precipitation forecasting using wavelet transform and neural networks,Qara-Qum catchment,Iran

This paper aims to provide a spatial and temporal analysis to prediction of monthly precipitation data which are measured at irregularly spaced synoptic stations at discrete time points. In the present study, the rainfall data were used which were observed at four stations over the Qara-Qum catchment, located in the northeast of Iran. Several models can be used to spatially and temporally predict the precipitation data. For temporal analysis, the wavelet transform with artificial neural network (WTANN) framework combines with the wavelet transform, and an artificial neural network (ANN) is used to analyze the nonstationary precipitation time-series. The time series of dew point, temperature, and wind speed are also considered as ancillary variables in temporal prediction. Furthermore, an artificial neural network model was used for comparing the results of the WTANN model. Therefore, four models were developed, including WTANN and ANN with and without ancillary data. Several statistical methods were used for comparing the results of the temporal analysis. It was evident that at three of the four stations, the WTANN models were more effective than the ANN models, and only at one station, the ANN model with ancillary data had better performance than the WTANN model without ancillary data. The values of correlation coefficient and RMSE for WTANN model with ancillary data for the validation period at Mashhad station which showed the best results were equal to 0.787 and 13.525 mm, respectively. Finally, an artificial neural network model was used as an alternative interpolating technique for spatial analysis.     

Study on the Reasonable Height of Embankment in Qinghai–Tibet Highway

A reasonable height of embankment is beneficial for maintaining the thermal and mechanical stability of highway in cold regions. This paper firstly introduced theoretical models for two main sources of settlement, including an improved consolidation theory for thawing permafrost and a simple rheological element based creep model for warm frozen soils. A modified numerical method for living calculating thaw consolidation and creep in corresponding domains and for post-processing the proportion of each source in total settlement based on the effective thaw consolidation time. Two typical geological sections underlain by warm permafrost layer were selected from the Qinghai–Tibet highway. The heat transfer and continuing settlement for two sections were modeled by assuming that the height of embankment ranges from 0 to 6.0 m. The reasonable critical height for two sections are 1.63 and 1.35 m, respectively, by comparing maximum thawing depth, mean annual temperature and settlement in the roadbed center. For two sections with design height of embankment, the proportions of thaw consolidation and creep to the total settlement were analyzed. For sections at higher ground temperature, thaw consolidation accounts for a major part while thaw consolidation of section L is a little larger than that of creep.     

http://www.sciencedirect.com/science/article/pii/S1674987114001376

A new model is derived to predict the peak ground acceleration(PGA) utilizing a hybrid method coupling artificial neural network(ANN) and simulated annealing(SA), called SA-ANN. The proposed model relates PGA to earthquake source to site distance, earthquake magnitude, average shear-wave velocity,faulting mechanisms, and focal depth. A database of strong ground-motion recordings of 36 earthquakes,which happened in Iran's tectonic regions, is used to establish the model. For more validity verification,the SA-ANN model is employed to predict the PGA of a part of the database beyond the training data domain. The proposed SA-ANN model is compared with the simple ANN in addition to 10 well-known models proposed in the literature. The proposed model performance is superior to the single ANN and other existing attenuation models. The SA-ANN model is highly correlated to the actual records(R=0.835 and r =0.0908) and it is subsequently converted into a tractable design equation.     

人工神经网络在海浪数值预报中的应用

探讨将人工神经网络技术和传统的数值模式相结合,以期得到一个更有效的海浪预报方法.以第3代海浪模式的模拟结果作为输入,浮标观测资料作为输出,采用人工神经网络进行训练,训练的初步结果显示,人工神经网络可以改进海浪数值模式的预报精度,但在波高比较大时,改进的效果并不令人满意.为此,对观测值大于1.5m时的有效波高进行再训练,从而结果有了进一步的改善.研究结果证明人工神经网络技术可以提高海浪数值预报的精度.     

人工神经网络模型在地学研究中的应用进展

近年来,随着人工神经网络（ANNs）自身技术的不断完善,应用ANNs模型成功解决各类地学问题的案例大量出现。通过对其发展历程进行分析发现,20世纪80年代末国际地学分析中已开始融入ANNs技术,国内则滞后 1～2年。在地学分析中使用的各类人工神经网络类型中,BP模型应用最广,占到85％以上。在10余年的应用过程中,虽然地学的各个分支学科都移植了一种或数种ANNs模型作为其分析工具,但水文、地质、大气、遥感等领域应用较为广泛。传统地学定量分析中的单变量或多变量预测成为人工神经网络地学模型的主要应用客体。同时,诸如模式识别和过程模拟等也是ANNs模型求解的对象。目前,随着建模经验和知识的积累,地学ANNs模型的发展呈现出多种技术综合集成的态势,遗传算法、小波转换、模拟退火算法以及模糊逻辑等方法与ANNs模型融合,成为解决地学分析中非线性问题的利器。     

GIS and ANN-based spatial prediction of DOC in river networks: a case study in Dongjiang, Southern China

This paper investigates the use of an artificial neural network (ANN) model to predict dissolved organic carbon (DOC) in a river network and evaluates the impacts of watershed characteristics on stream DOC. Samples and relevant environmental variables were obtained from field sampling at 28 hydrological response units (HRUs) and a MODIS/SRTM DEM satellite image. HRUs can provide reliable spatial interpolation for filling data gaps and incorporate potential spatial correlation among observations in each ANN neuron. The process and results of neural network modeling were assessed by deterministic and statistical methods and spatial regression kriging. The spatial prediction results show that ANN, using improved back propagation algorithms of 7-15-1 architecture, was the optimal network, by which predictions maintained most of the original spatial variation and eliminated smoothing effects of RK. The sum of the relative contributions of four sensitive variables, including soil organic carbon density, geographic longitude, surface runoff and Chl a in river water, was >75 %. A minor prediction error of ~6 % was found in HRUs of open shrublands, but HRUs of urban and croplands had an error of 24–30 %. This pattern exemplifies anthropogenic impacts in urban areas on stream DOC and agricultural activities in croplands. The usefulness of ANN modeling-based GIS in this study is demonstrated by depiction of spatial variation of stream DOC and indicates the benefits of understanding sensitive factors for watershed impact assessments.     

人工神经网络方法在径流预报中的应用

采用ＢＰ神经网络模型,以西北内陆河黑河流莺峡年平均出山地表流量为研究对象,对人工神经网络研究方法在干旱区环流径流预报中的应用进行了初步尝试,结果表明该预报成功率较高,证实了人工神经网络方法应用于流量预报领域的可行性,并分析了该方法在预报过程中的优缺性。     

Rockfall hazard assessment at Ajanta Cave,Aurangabad, Maharashtra,India

The Ajanta caves are situated in Deccan Trap basalt and declared as one of the World Heritage Sites by UNESCO. The present study aims to investigate and understand the damage of caves and to protect the life of the visitors from the rockfall phenomenon at and around the caves. Information related to the detached rock mass/block was acquired by using Barton–Bandis model in Universal Distinct Element Code. Parameters for rockfall simulation were determined by rigorous field study and laboratory experiment and then calibrated some of the parameters by back analysis. RocFall 4.0 program has been used to calculate maximum bounce heights, total kinetic energies, and translational velocities of the falling blocks of different weights. The maximum bounce height varies from 14.0 to 19.0 m for the weight of the block size ranging from 500 to 2,000 kg, whereas the maximum velocity and maximum kinetic energy are 30.0 m/s and 917.66 kJ, respectively. Finally, the results of simulation have been used to find out the position of the barrier and its capacity to design the protection barrier. The barrier capacity was found to be 325 kJ for 2,000 kg of falling blocks at a height of 50.0 m.     

Artificial Neural Network as a Tool for Backbreak Prediction

Backbreak is one of the destructive side effects of the blasting operation. Reducing of this event is very important for economic of a mining project. Involvement of various parameters has made the backbreak analyzing difficult. Currently there is no any specific method to predict or control the phenomenon considering all the effective parameters. In this paper, artificial neural network (ANN) as a powerful tool for solving such complicated problems is used to predict backbreak in blasting operation of the Sangan iron mine, Iran. Network training was fulfilled using a collected database of the practiced operation including blast design details and rock condition. Trying various types of the networks, a network with two hidden layers was found to be optimum. Performance of the ANN model was compared with statistical analysis using datasets which were kept apart from the original database. According to the obtained results, for the ANN model there existed a higher correlation (R² = 0.868) and lesser error (RMSE = 0.495) between the predicted and measured backbreak as compared to the regression model. Also, sensitivity analysis revealed that the inputs rock factor and number of rows are the most and the least sensitive parameters on the output backbreak, respectively.     

Landslide susceptibility mapping at Vaz Watershed (Iran) using an artificial neural network model: a comparison between multilayer perceptron (MLP) and radial basic function (RBF) algorithms

Landslide susceptibility and hazard assessments are the most important steps in landslide risk mapping. The main objective of this study was to investigate and compare the results of two artificial neural network (ANN) algorithms, i.e., multilayer perceptron (MLP) and radial basic function (RBF) for spatial prediction of landslide susceptibility in Vaz Watershed, Iran. At first, landslide locations were identified by aerial photographs and field surveys, and a total of 136 landside locations were constructed from various sources. Then the landslide inventory map was randomly split into a training dataset 70 % (95 landslide locations) for training the ANN model and the remaining 30 % (41 landslides locations) was used for validation purpose. Nine landslide conditioning factors such as slope, slope aspect, altitude, land use, lithology, distance from rivers, distance from roads, distance from faults, and rainfall were constructed in geographical information system. In this study, both MLP and RBF algorithms were used in artificial neural network model. The results showed that MLP with Broyden–Fletcher–Goldfarb–Shanno learning algorithm is more efficient than RBF in landslide susceptibility mapping for the study area. Finally the landslide susceptibility maps were validated using the validation data (i.e., 30 % landslide location data that was not used during the model construction) using area under the curve (AUC) method. The success rate curve showed that the area under the curve for RBF and MLP was 0.9085 (90.85 %) and 0.9193 (91.93 %) accuracy, respectively. Similarly, the validation result showed that the area under the curve for MLP and RBF models were 0.881 (88.1 %) and 0.8724 (87.24 %), respectively. The results of this study showed that landslide susceptibility mapping in the Vaz Watershed of Iran using the ANN approach is viable and can be used for land use planning.     

Burden prediction in blasting operation using rock geomechanical properties

Burden prediction is a vital task in the production blasting. Both the excessive and insufficient burden can significantly affect the result of blasting operation. The burden which is determined by empirical models is often inaccurate and needs to be adjusted experimentally. In this paper, an attempt was made to develop an artificial neural network (ANN) in order to predict burden in the blasting operation of the Mouteh gold mine, using considering geomechanical properties of rocks as input parameters. As such here, network inputs consist of blastability index (BI), rock quality designation (RQD), unconfined compressive strength (UCS), density, and cohesive strength. To make a database (including 95 datasets), rock samples are used from Iran’s Mouteh goldmine. Trying various types of the networks, a neural network, with architecture 5-15-10-1, was found to be optimum. Superiority of ANN over regression model is proved by calculating. To compare the performance of the ANN modeling with that of multivariable regression analysis (MVRA), mean absolute error (E _a), mean relative error (E _r), and determination coefficient (R ²) between predicted and real values were calculated for both the models. It was observed that the ANN prediction capability is better than that of MVRA. The absolute and relative errors for the ANN model were calculated 0.05 m and 3.85%, respectively, whereas for the regression analysis, these errors were computed 0.11 m and 5.63%, respectively. Moreover, determination coefficient of the ANN model and MVRA were determined 0.987 and 0.924, respectively. Further, a sensitivity analysis shows that while BI and RQD were recognized as the most sensitive and effective parameters, cohesive strength is considered as the least sensitive input parameters on the ANN model output effective on the proposed (burden).     

人工神经网络在遥感图像处理中的应用探讨

遥感图像处理常见的困难有数据量巨大、噪声信息多，高度非线性及其导致的难以用解析或表述处理模型等。人工神经网络（artificial neural network,ANN)是由大量简单神经元广泛相互联接而成的非线性映射或自适应动力系统，可以解决上述问题，使用ANN进行遥感图像处理在遥感图像复原，变换和分类中有如下应用：（1）使用ANN和必要辅助数据从TM图像中提取地下火热辐射数据；（2）构造ANN非线性映射，利用TM1－5，7图像提高TM6图像空间分辨率；（3）模糊神经网络（FNN）遥感图像分类。     

Mechanisms of the development of water-conducting fracture zone in overlying strata during shortwall block backfill mining: a case study in Northwestern China

In this work, a shortwall block backfill mining (SBBM) technique is proposed for the recovery of residual corner coal pillars and irregular blocks left behind during the exploitation of coal mines, and a solution is provided for the risks associated with gangue piling and the loss of water resources owing to coal mining. Based on the theory of beams on elastic foundations, a mechanical analysis model was established for calculating the height of a water-conducting fracture zone (WCFZ) in the overlying strata of coal mines exploited using the SBBM technique. It was found that the key factors influencing the development of the WCFZ are the mining height, width of the protective coal pillars, backfill percentage, block length, and number of mining blocks. The relationships between these factors and the height of the WCFZ were obtained by incorporating the relevant parameters in the above-mentioned model. In the field experiment site, it was discovered that the minimum coal pillar width and goaf backfill percentage required to prevent the development of water-conducting fractures that could reach an aquifer are 5 m and 65%, respectively. Based on this result, the protective pillars of the site were designed to be 5 m wide, while the goaf backfill percentage was set as 80%. The borehole fluid method was used to measure the height of the WCFZ, which was found to be 26.8 m. This is consistent with the theoretical calculations (27.0 m) of this study, and thus, validates the reliability of the proposed mechanical model. The findings of this work will improve the recovery rate of residual coal resources in coal mining areas, and they are significant for the refinement of water conservation mining theories.     

Predicting elastic properties of schistose rocks from unconfined strength using intelligent approach

Elastic properties of rocks play a major and crucial role for the design of any engineering structure. Determination of elastic properties in laboratory is tedious, laborious, very time consuming, as well as expertise is required, whereas determination of uniaxial compressive strength (UCS) and tensile strength in laboratory is simple, easy, and less expertise is required. Here, an attempt has been made to predict the elastic properties (Poisson’s ratio and Young’s modulus) of the schistose rocks from unconfined strength (UCS and tensile strength) using artificial neural network (ANN). A three-layer feed-forward back propagation neural network with 2-5-2 architecture was trained up to 855 epochs to predict the elastic properties of rock mass. The network was trained and tested by 120 data sets, and validation of the network was done by 20 new randomly selected data sets of UCS and tensile strength. The samples were collected from the schistose rocks of Nathpa-Jhakri hydropower project site, SJVNL, Himachal Pradesh, India. To check the validity and suitability of the artificial neural network technique, multivariate regression analysis (MVRA) is also performed, and comparison has been made. It was found that ANN gives closer values of predicted Poisson’s ratio and Young’s modulus as compared to MVRA. The coefficient of determination for Poisson’s ratio was 0.9809 and 0.843 by ANN and MVRA, respectively, whereas 0.9922 and 0.9362 for Young’s modulus by ANN and MVRA, respectively. The mean absolute percentage error (MAPE) for Young’s modulus is 11.13 and 28.21 by ANN and MVRA, respectively; whereas MAPE for Poisson’s ratio is 3.64 and 9.23 by ANN and MVRA, respectively.     

Prediction and controlling of flyrock in blasting operation using artificial neural network

Flyrock is one of the most hazardous events in blasting operation of surface mines. There are several empirical methods to predict flyrock. Low performance of such models is due to complexity of flyrock analysis. Existence of various effective parameters and their unknown relationships are the main reasons for inaccuracy of the empirical models. Presently, application of new approaches such as artificial intelligence is highly recommended. In this paper, an attempt has been made to predict and control flyrock in blasting operation of Sangan iron mine, Iran incorporating rock properties and blast design parameters using artificial neural network (ANN) method. A three-layer feedforward back-propagation neural network having 13 hidden neurons with nine input parameters and one output parameter were trained using 192 experimental blast datasets. It was also observed that in ascending order, blastability index, charge per delay, hole diameter, stemming length, powder factor are the most effective parameters on the flyrock. Reducing charge per delay caused significant reduction in the flyrock from 165 to 25 m in the Sangan iron mine.     

Comparison of cokriging and adaptive neuro-fuzzy inference system models for suspended sediment load forecasting

The present research was carried out by using artificial neural network (ANN), adaptive neuro-fuzzy inference system (ANFIS), cokriging (CK) and ordinary kriging (OK) using the rainfall and streamflow data for suspended sediment load forecasting. For this reason, the time series of daily rainfall (mm), streamflow (m³/s), and suspended sediment load (tons/day) data were used from the Kojor forest watershed near the Caspian Sea between 28 October 2007 and 21 September 2010 (776 days). Root mean square error, efficiency coefficient, mean absolute error, and mean relative error statistics are used for evaluating the accuracy of the ANN, ANFIS, CK, and OK models. In the first part of the study, various combinations of current daily rainfall, streamflow and past daily rainfall, streamflow data are used as inputs to the neural network and neuro-fuzzy computing technique so as to estimate current suspended sediment. Also, the accuracy of the ANN and ANFIS models are compared together in suspended sediment load forecasting. Comparison results reveal that the ANFIS model provided better estimation than the ANN model. In the second part of the study, the ANN and ANFIS models are compared with OK and CK. The comparison results reveal that CK was a better estimation than the OK. The ANFIS and ANN models also provided better estimation than the OK and CK models.